The most useful thing your company knows was said out loud, and then lost.
Everything an agent needs to be useful about your work has already been discussed: why the architecture is like that, what the customer actually objected to, what was decided and quietly reversed. It was in a meeting. Nobody wrote it down, and the recording that did capture it sits in a vendor's cloud where nothing can read it.
Ledgeur was built to record meetings on your own machine. That is still what it does, and it is still the reason to install it. But a record nothing can read is an archive, and an archive is a place things go to stop being useful.
So the record is open by design. Everything the app holds is reachable over the Model Context Protocol by the agents you already use, running as you, under the same row-level security that governs what you can see. An agent cannot read a meeting you could not.
5 tools, and no scraping.
This list is generated from the code that implements it, on this page and on the agents page, so it cannot describe a tool that does not exist.
list_meetingsList the user's most recent meetings (title, date, status). Use to browse before drilling into one.search_meetingsSearch the user's meetings by keyword in the title. Returns matching meetings.get_meetingGet a single meeting by id: its summary, decisions, open questions, the full Markdown notes, the whole transcript as text, and the speakers and timed segments behind it.list_peopleList the people who appear across the user's meetings, with how many meetings each was in and which ones. Only people who have been named — unnamed voices are not people.list_tasksList action items (tasks) extracted from meetings, optionally filtered by status.Same tools whether the agent connects over stdio on your own machine or to the hosted endpoint. How to connect one.
And when meetings should feed the whole company's memory.
Contextely is a context layer: it condenses what a company knows across Notion, Drive, Postgres and any MCP source into one searchable memory, scored by what the person asking is entitled to see. Ledgeur connects to it in both directions.
Paste a personal Contextely key into Settings and asking Ledgeur a question reaches the company’s memory alongside your own meetings. The key is validated and kept server-side, never on the device, the same as every other integration’s secret.
A Contextely admin adds Ledgeur as a source and it ingests over MCP, using the same list_meetings and get_meeting tools any agent uses. Nothing is pushed from Ledgeur; Contextely pulls and refreshes on its own schedule, under its own entitlement rules.
Contextely is a separate product with its own pricing, and Ledgeur works perfectly well without it. It is listed here because “meetings become memory a company can query” is the whole point of this page, and that integration is the shipped version of it rather than a description of one.
Why this is a different bet from everyone else's.
The whole category is heading toward the same idea. The disagreement is about where the memory lives.
The well-funded version of this is a vendor building a proprietary index of your company’s conversations in their cloud, and selling you access to it. It works, and it produces exactly one company that can read your meetings and one company that can stop.
Ledgeur’s version keeps the transcription on your machine, the source under an MIT licence, the schema in a repository you can read, and the interface an open protocol rather than a private API. If the company disappears, the record keeps working, because it is local files and a Postgres schema you already have.
That is not a better product by every measure and it is not meant to be. It is a different answer to the question of who owns the context, and it is the one worth building for a company that intends to still be able to read its own history in five years.
Start with one meeting
Record it, read it, then point Claude at it. The recording part is free forever; agent access is part of the Team plan.